Country evidence / Definitions and sources
About the atlas
The measures cover different samples and periods. Missing values are shown separately from zero.
Paper 2 / Founded 2015–2022
AI startups
Startups founded in 2015–2022 with an operating country and an AI classification in the paper's reconciled PitchBook–Stears sample. These are sample counts, not a census of all startups. Firms developing or selling AI products, services, infrastructure or compute. Internal use of AI alone does not qualify. Current-description AI status is held fixed across financing years. Operating country follows the adopted early-worker-location and company-record assignment, not simply legal incorporation or headquarters.
The African sample contains 1,128 AI startups among 6,125 startups in 47 countries. A country without sampled firms is marked as not covered; that does not establish that it has no startups.
Paper 2 / Financing 2019–2025
AI startup funding ranks
Sum of observed annual financing in 2019–2025, in nominal US dollars, including equity, debt, grants and other financing. Uses adopted deduplicated event, tranche-date and recipient rules. Excludes post-money valuations and investor-level imputed allocations. Years with no financing event enter as observed zero under the paper's rules. Unresolved amounts remain missing and contribute nothing to the displayed observed sum; they are counted in missing firm-years. Partial sums are not estimates of complete national funding. Country ranks use the full adopted Paper 2 descriptive sample of 116,795 startups across 172 operating countries, before regression-support exclusions. Funding ranks compare observed AI-producer financing in 2019–2025, summed in nominal US dollars; startup ranks compare AI-producer counts among firms founded in 2015–2022. Positive country totals rank from largest to smallest using competition ties: equal totals share a rank and the following rank skips those positions. Observed zero has rank 0 (Unranked); missing funding or no country sample has rank null. Africa ranks use countries in the existing 54-country atlas roster; seven have no sample. Ranks compare recorded sample totals, not a census of national activity. Partial funding sums retain their missing firm-years and are not complete national totals.
Africa and global ranks compare countries by recorded AI startup funding in the same study sample. Equal totals share a rank; the next rank skips the tied positions. Countries with zero recorded AI funding are unranked. Missing coverage is shown separately.
Funding totals cover the full study sample, before the paper’s estimation exclusions. Coverage can differ between AI and non-AI firms.
An observed zero means no funding under the paper’s event-accounting rules. If no amount can be established, the value remains missing.
Country workforce estimates / 2023
Share of Africa’s AI-skilled workers
Each country’s corrected, coverage-adjusted AI-worker estimate divided by the sum of nonmissing estimates across African countries, multiplied by 100. Employed people with a completed bachelor degree or higher at December 31, 2023, across all employers and industries. AI skill requires a finite strict5 relatedness score of at least 0.55 and a recorded AI-acquisition date on or before the snapshot. One person-year is split equally across distinct concurrent workplace countries, including unknown locations, and deduplicated across employers. Raw country masses receive the fixed adopted GSJ country coverage factor exactly once; JWB person-history weights are not applied. Missing skill scores remain in the BA+ denominator; qualifying scores without a dated onset are reported separately. These are coverage-adjusted profile-based estimates, not census counts or AI-tool-use measures.
Shares sum to 100% across 54 countries with classified BA+ employment support. Countries without support are excluded, not treated as zero.
Country coverage factors retain their original benchmark reference years.
RADIA · OpenAlex / Historical research, 2011–2015
University AI research
Rankings use AI-focused researchers affiliated with universities in 2011–2015, measured continuously rather than using the AI-strong cutoff. Across an author's university-affiliated papers in the fixed window, AI plus non-AI computer science comprises at least one third of all papers; AI comprises at least half of that AI-plus-CS portfolio. The adopted primary topic definition is unchanged. Adopted OpenAlex educational-institution registry, including reviewed historical units. Observed researcher coverage, completed download and adopted historical-identity exclusions determine eligibility. Not restricted to MNE-linked institutions or the binary AI-strong subset.
Researcher counts refer to authors with recorded affiliations, not verified faculty or workforce headcounts. An author may appear at more than one institution. Unknown strength is not scored as zero.
Country ranks use the sum of AI-focused researcher counts across eligible universities. These are researcher–university affiliations: a researcher affiliated with two universities can contribute to both. Individual universities are ranked by their own AI-focused researcher counts. Equal values share a rank.
Nigeria
987 AI-focused researcher affiliations across 164 scored institutional records.